|Publication number||US5717926 A|
|Application number||US 08/704,336|
|Publication date||Feb 10, 1998|
|Filing date||Aug 28, 1996|
|Priority date||Feb 28, 1995|
|Publication number||08704336, 704336, US 5717926 A, US 5717926A, US-A-5717926, US5717926 A, US5717926A|
|Inventors||Luke Matthew Browning, Olivier Alain Denis Cremel, Jeffrey Scott Peek|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (8), Non-Patent Citations (2), Referenced by (39), Classifications (5), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This is a continuation of application Ser. No. 08/396,309 filed Feb. 28, 1995, now abandoned.
1. Field of the Invention
The present invention relates to operating system environments and, more particularly, to efficient forking of a process.
2. Background Information and Description of the Related Art
For background purposes, this section includes a general description of a conventional UNIX™-based computing system.
Virtual Memory: Conventional UNIX-based computing systems typically utilize virtual memory. Virtual memory extends the effective size of the computing system's real memory by utilizing disk space to simulate additional real memory. The computer system's operating system may divide virtual memory into large units, called segments (e.g. 256M bytes each), and then further divide those segments and real memory into smaller units, called pages (e.g. 4K bytes per page).
Critical Sections: Conventional UNIX-based computing systems typically include "critical sections" of code and/or shared data structures, such as shared libraries, whose integrity is extremely important for correct system operation. Operating systems may incorporate software blocking and/or read-only locks to "serialize" access to those critical sections. Blocking locks cause a thread (described herein) requesting the critical sections to cease executing, i.e., to go to "sleep", if another thread currently holds the lock. However, as can be seen, the disadvantage of blocking locks is that all other threads may be suspended when the critical section is executing.
Virtual Memory Manager: Conventional UNIX-based operating systems may include a "virtual memory manager" (VMM) for tracking the addresses of pages residing in real memory (i.e. real addresses) and the virtual addresses of pages that must be retrieved from disk when referenced. To do this, the VMM creates a descriptor table for mapping virtual addresses to real addresses. However, when a virtual address does not directly map to a real address, a page fault occurs. In this case, the VMM moves page(s) in real memory to disk, retrieves the desired page(s) from disk or real memory, and stores those retrieved pages in the location of the previously moved page(s).
Pinned and Virtually Pinned Pages: A "pinned" page is a page in real memory that is flagged so that the VMM cannot move it. As such, a page fault cannot occur if the system references a pinned page. Another technique to help insure that pages will be located in real memory is to "virtually" pin those pages. To describe virtual pinning, it is important to know that when a page fault occurs, the VMM moves the least recently referenced page(s) to disk. Therefore, if the OS "touches" (i.e. references) virtual pages in a critical section, it is likely that a page fault will not occur in the critical section because those pages were recently referenced.
Processes and Threads: Conventional UNIX™-based operating systems schedule execution of various user processes on one or more microprocessors. A user process includes a set of system resources for executing one or more threads. A thread is a simple execution path through application software and the operating system. Each thread has its own kernel stack for executing system calls and kernel services on behalf of the thread. The kernel stack contains the thread's processing information, such as the automatic variable, hardware registers, and a backlink pointer to the previous stack frame. The set of system resources for the process includes code, data, each thread's kernel stack, hardware context, and process tables, all collectively referred to as "process image." Conventionally, the operating system groups most, if not all, process image information, including all kernel stacks, in one memory location (e.g., segment) so that they may be easily copied.
Fork System Call (Forkcopy): A calling thread in a user process (herein "parent process") can create a new user process (herein "child process") by initiating a fork system call. Performing the fork is rather straight forward if the parent process is single-threaded. For example, referring to FIG. 1, the OS forkcopies the parent's process image contained in kernel space 20 and user space 10 (both located in seg. 2) to create the new child process image. As a result, the operating system forkcopies seg. 2 in its entirety. However, if the operating system only permits one-to-one mapping of virtual addresses, the operating system may create a "dummy" parent. A dummy parent contains a mapping to all non-pinned pages of the parent process, having pointers to those pages by both the parent process and the child process.
Forking of a Multi-Threaded Process: Unfortunately, specific problems and disadvantages arise when a thread of a multi-threaded process attempts to initiate the fork system call. Importantly, each child process is single-threaded, even though the parent process may be multi-threaded. Accordingly, the child process inherits only the contents of the calling thread's kernel stack, i.e., it inherits no other thread's kernel stack. However, because conventional UNIX-based operating systems locate all kernel stacks of a single process in one memory location (e.g., segment), all kernel stacks of that process are forkcopied, causing several disadvantages and problems.
First, because the child process only inherits the calling thread's kernel stack, the most obvious disadvantage is that additional memory must be forkcopied.
Second, one of the non-calling threads running on its stack may be in a "disabled critical section." A disabled critical section provides a high level of serialization because the operating system disables all hardware interrupts and, therefore, page faults cannot occur. As a result, a disabled critical section must always reference pinned pages within real memory. Otherwise, the computer system could crash. However, as previously described above, the act of forkcopying a segment results in the remapping of the pages in that segment. Pages may be made read-only and associated with a dummy parent segment during a forkcopy. Consequently, if the thread in the disabled critical section references its stack, a page fault will occur.
Third, because date integrity of a disabled critical section is paramount, all referenced pages must be valid. Even if the system does not crash as a result of a page fault, the thread will be put to sleep, resulting in a loss of serialization. Another thread will be scheduled to run on the processor and this thread might reference the same data. As a result, the data could be in an inconsistent state if it was only partially updated.
Accordingly, there is a great demand for an enhanced operating system that permits a process (simultaneously executing multiple threads) to perform a forkcopy, while maintaining data integrity. Furthermore, it would be advantageous for this enhanced operating system to permit a single threaded process to perform a forkcopy without any performance degradation.
Accordingly, a computer-implemented method, computer system, and memory perform an efficient fork operation of a parent process. The parent process includes one or more threads, each having a kernel stack. The method includes the steps of locating the kernel stacks in a first memory location (e.g., segment 1) if the parent process is multi-threaded, in response to an initiation of a fork by a calling thread, switching the kernel stack of the calling thread to a second memory location if the parent process is multi-threaded, and copying the second memory location (e.g., segment 2) to create a child process.
Therefore, it is an object of the present invention to provide a forking technique that does not clone a memory segment containing any kernel stack other than the calling kernel stack.
It is another object of the present invention to reduce the amount of memory that is forkcopied.
It is a further object of the present invention to optimize the forking procedure for both single and multi-threaded parent processes.
It is another object of the invention to provide a forking technique that does not suspend other threads when the calling process is forking.
It is still another object of the present invention to preserve data integrity of threads in the process that are executing concurrent to the thread executing the fork system call.
These and other objects, features, and advantages will become even more apparent in light of the following detailed description and drawings.
FIG. 1 illustrates a prior art block diagram of virtual memory.
FIG. 2 illustrates a suitable hardware configuration for implementing the present invention.
FIG. 3 illustrates detailed logic for creating an additional thread in a process.
FIG. 4 illustrates detailed logic for performing a fork operation.
The following detailed description is of the best presently contemplated mode for carrying out the invention. This detailed description is not to be viewed in a limiting sense, but is made merely for the purpose of illustrating the general principles of the invention, the scope of which is only defined by the appended claims.
The preferred embodiment of the present invention includes a system, method, and memory for efficiently creating a process from a multi-threaded parent process. The preferred embodiment forms part of an enhanced UNIX-based operating system (OS) in the management and direction of workstation operations. More specifically, the preferred embodiment is implemented within the AIX™ operating system. However, one skilled in the art will readily recognize that the present invention could be implemented in any suitable operating system.
The present invention may be practiced in any suitable hardware configuration, such as the hardware configuration illustrated in FIG. 2, or, alternatively, a laptop computer. Referring to FIG. 2, workstation 200 includes any suitable central processing unit 210, such as a conventional microprocessor (e. g., Intel's™ Pentium™ or IBM's PowerPCT™), and a number of other units TM interconnected via system bus 212. Illustratively, workstation 200 includes random access memory ("RAM") 214, cache 215, non-volatile RAM 232, read only memory ("ROM") 216, display adapter 236 for connecting system bus 212 to display device 238, and I/O adapter 218 for connecting peripheral devices (e.g. disk and tape drives 220) to system bus 212. Workstation 200 further includes user interface adapter 222 for connecting keyboard 224, mouse 226, speaker 228, microphone 232, and/or other user interface devices, such as a touch screen device (not shown), to system bus 212. Communication adapter 234 connects workstation 200 to a data processing network.
The enhanced OS, as computer readable program code, resides within a machine-readable media (e.g., memory) to direct the operation of workstation 200. Any suitable memory may retain the enhanced OS, such as RAM 214, ROM 216, a magnetic diskette (e.g., floppy diskette), CD-ROM, magnetic tape, or optical disk (the last four being located in disk and tape drives 220). While the specification describes the OS as performing the features of the invention, in actuality, the OS merely provides the instructions for the microprocessor to perform those features.
FIG. 3 illustrates detailed logic for creating an additional thread in a process. Referring to FIG. 3, at 310, the user process calls a thread library. This library contains a library of functions and is an extension of the kernel. For example, the library transfers user system calls into an appropriate sequence of kernel system calls.
In turn, at 320, the thread library issues a thread-create system call. In response, at 330, the system call handler examines a current process data structure in segment 14 (see 40 in FIG. 1) to determine if segment 1 (see 5 in FIG. 1) exists. The current process data structure contains segment definitions and information common to all thread(s) of the current process. If segment 1 exists, the system call handler maps segment 1 into memory. Next, the system call handler examines the current thread data structure located in segment 14 and switches to a kernel stack identified by that data structure. Therefore, the system call handler switches from a user stack defined by the process to the identified kernel stack and uses it for subsequent calls in the kernel. If the current process is single threaded, the current thread's kernel stack is located in kernel space 20 of segment 2, which is the default kernel stack. If the current process is multi-threaded, the current thread's kernel stack will be located in segment 1 (described herein). Next, the system call handler issues a thread create call.
At 340, the kernel examines the current process data structure in segment 14 to determine if a kernel stack segment (i.e., segment 1) exists for the current process. As previously described, the current process data structure contains segment definitions of the current process. Therefore, if the kernel stack segment does not exist, the second thread of the process is being created. That is, the current process is single threaded. Control is then directed to 350. However, if the kernel stack segment does exist, control is directed to 360.
At 350, the kernel creates a kernel stack segment (e.g., seg. 1) for the current process and allocates a kernel stack in the kernel stack segment for the current thread. The kernel then sets a kernel stack pointer in segment 14 to the kernel stack for the current thread. Accordingly, each thread has a defined kernel stack location in the kernel stack segment. This setting does not take affect until the next system call.
At 360, the kernel creates a new thread in the current process by copying the calling thread data structure (i.e., the data structure defining the calling thread, located in segment 14) in its entirety and assigns it a kernel stack in the kernel stack segment (e.g., segment 1). The kernel then returns control to the caller (i.e. user process). Accordingly, the default kernel stack in kernel space 20 of segment 2 is free.
FIG. 4 illustrates detailed logic for performing a fork operation of a current (i.e., calling) process. The current process may be multi-threaded or single threaded. At 410, the process decides to initiate forking of a new child process. At 420, the process calls a thread library to prepare for a fork. This thread library contains a library of functions and is an extension of the kernel. The process issues the KFORK system call to transfer the fork system call into an appropriate sequence of kernel system calls.
At 430, the system call handler examines the current process data structure in segment 14 to determine if segment 1 exist. If segment 1 exists, the system call handler maps segment 1 into memory. Next, the system call handler examines the current thread data structure located in segment 14 and switches to a kernel stack identified by that data structure. Therefore, the system call handler switches from the user stack defined by the process to the identified kernel stack and uses it for subsequent calls in the kernel. Again, if the current process is single threaded, the current thread's kernel stack is located in segment 2, which is the default kernel stack. If the current process is multi-threaded, the current thread's kernel stack will be located in segment 1. The system call handler then calls the KFORK system call.
At 440, the kernel determines if the calling process is multi-threaded. To do this, the kernel examines a count in the calling process data structure located in segment 14. If the process is multi-threaded, at 450, the kernel checks a flag, called SFORKSTACK, in the current process structure found in segment 14. If this flag has been set, at 460, the thread sleeps on a process-based event list. That is, the thread will cease processing until the SFORKSTACK flag clears. Control then returns to 440. The calling thread resumes from sleep when another thread finishes executing either a fork, exit, or exec call.
Returning to 450, if the SFORKSTACK flag was not set, at 470, the kernel sets the SFORKSTACK flag and switches kernel stacks from segment 1 to the location of the previously empty default kernel stack in segment 2. At 480, the kernel creates the child process by copying segment 2 in its entirety. The kernel also initializes the child thread to begin running and, at 490, switches the kernel stack back to segment 1, clears the SFORKSTACK flag, and wakes up any sleepers on the process based list. Control then returns to the user mode.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4812981 *||Oct 24, 1985||Mar 14, 1989||Prime Computer, Inc.||Memory management system improving the efficiency of fork operations|
|US4849877 *||Dec 22, 1986||Jul 18, 1989||American Telephone And Telegraph Company||Virtual execution of programs on a multiprocessor system|
|US4901231 *||Dec 22, 1986||Feb 13, 1990||American Telephone And Telegraph Company||Extended process for a multiprocessor system|
|US5159687 *||Dec 27, 1989||Oct 27, 1992||Caseworks, Inc.||Method and apparatus for generating program code files|
|US5247673 *||Nov 16, 1990||Sep 21, 1993||Bull Hn Information Systems Italia S.P.A.||Multiprocessor system having distributed shared resources and dynamic global data replication|
|US5313647 *||Sep 20, 1991||May 17, 1994||Kendall Square Research Corporation||Digital data processor with improved checkpointing and forking|
|US5339415 *||Nov 9, 1992||Aug 16, 1994||Cray Research, Inc.||Dual level scheduling of processes to multiple parallel regions of a multi-threaded program on a tightly coupled multiprocessor computer system|
|US5404521 *||Oct 5, 1993||Apr 4, 1995||Top Level Inc.||Opportunistic task threading in a shared-memory, multi-processor computer system|
|1||"Fork Clone Address Space Implementation on MVS", IBM Technical Disclosure Bulletin, vol. 35, No. 6, Nov. 1992, pp. 363-367.|
|2||*||Fork Clone Address Space Implementation on MVS , IBM Technical Disclosure Bulletin, vol. 35, No. 6, Nov. 1992, pp. 363 367.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US5974438 *||Dec 31, 1996||Oct 26, 1999||Compaq Computer Corporation||Scoreboard for cached multi-thread processes|
|US5974536 *||Aug 14, 1997||Oct 26, 1999||Silicon Graphics, Inc.||Method, system and computer program product for profiling thread virtual memory accesses|
|US6044445 *||Jun 28, 1996||Mar 28, 2000||Kabushiki Kaisha Toshiba||Data transfer method and memory management system utilizing access control information to change mapping between physical and virtual pages for improved data transfer efficiency|
|US6052708 *||Oct 23, 1997||Apr 18, 2000||International Business Machines Corporation||Performance monitoring of thread switch events in a multithreaded processor|
|US6138152 *||Jun 2, 1998||Oct 24, 2000||Lucent Technologies Inc.||Technique for effectively serving information objects through a communication network|
|US6141793 *||Apr 1, 1998||Oct 31, 2000||Hewlett-Packard Company||Apparatus and method for increasing the performance of interpreted programs running on a server|
|US6167437 *||Sep 2, 1997||Dec 26, 2000||Silicon Graphics, Inc.||Method, system, and computer program product for page replication in a non-uniform memory access system|
|US6173309 *||Feb 20, 1997||Jan 9, 2001||Hewlett-Packard Company||Null thread library and thread abstraction interface|
|US6249802||Sep 19, 1997||Jun 19, 2001||Silicon Graphics, Inc.||Method, system, and computer program product for allocating physical memory in a distributed shared memory network|
|US6269391 *||Jun 19, 1997||Jul 31, 2001||Novell, Inc.||Multi-processor scheduling kernel|
|US6289424||Sep 19, 1997||Sep 11, 2001||Silicon Graphics, Inc.||Method, system and computer program product for managing memory in a non-uniform memory access system|
|US6336177||Oct 13, 1999||Jan 1, 2002||Silicon Graphics, Inc.||Method, system and computer program product for managing memory in a non-uniform memory access system|
|US6418460 *||Feb 18, 1997||Jul 9, 2002||Silicon Graphics, Inc.||System and method for finding preempted threads in a multi-threaded application|
|US6654781 *||Dec 11, 1998||Nov 25, 2003||International Business Machines Corporation||Enhanced thread processing|
|US6687812 *||Apr 20, 2000||Feb 3, 2004||Nec Corporation||Parallel processing apparatus|
|US6708194 *||Apr 26, 2001||Mar 16, 2004||At&T Corp.||Porting POSIX-conforming operating systems to Win32 API-conforming operating systems|
|US6907517||Jun 7, 2002||Jun 14, 2005||Nec Corporation||Interprocessor register succession method and device therefor|
|US7139832 *||Aug 29, 2001||Nov 21, 2006||Hitachi, Ltd.||Data transfer and intermission between parent and child process|
|US7243345 *||Jul 8, 2002||Jul 10, 2007||Nec Corporation||Multi-thread executing method and parallel processing system|
|US7389297 *||Sep 2, 2004||Jun 17, 2008||Sun Microsystems, Inc.||Method for exit negotiation with aggregate application descendents|
|US7552434 *||Apr 30, 2004||Jun 23, 2009||Hewlett-Packard Development Company, L.P.||Method of performing kernel task upon initial execution of process at user level|
|US7587725 *||Jan 24, 2005||Sep 8, 2009||Microsoft Corporation||Methods and systems for creating and communicating with computer processes on remote systems|
|US7665143 *||May 16, 2005||Feb 16, 2010||Microsoft Corporation||Creating secure process objects|
|US7827380 *||Nov 4, 2008||Nov 2, 2010||International Business Machines Corporation||Method and apparatus for supporting shared library text replication across a fork system call|
|US9189605||Feb 23, 2009||Nov 17, 2015||Microsoft Technology Licensing, Llc||Protected computing environment|
|US20020065934 *||Aug 29, 2001||May 30, 2002||Hitachi, Ltd.||Data transfer method|
|US20030014473 *||Jul 8, 2002||Jan 16, 2003||Nec Corporation||Multi-thread executing method and parallel processing system|
|US20030028755 *||Jun 7, 2002||Feb 6, 2003||Nec Corporation||Interprocessor register succession method and device therefor|
|US20050138624 *||Jan 24, 2005||Jun 23, 2005||Microsoft Corporation||Methods and systems for creating and communicating with computer processes|
|US20050251803 *||Apr 30, 2004||Nov 10, 2005||Turner Yoshio F||Method of performing kernel task upon initial execution of process at user level|
|US20060242406 *||Apr 27, 2005||Oct 26, 2006||Microsoft Corporation||Protected computing environment|
|US20060248594 *||Apr 27, 2005||Nov 2, 2006||Microsoft Corporation||Protected media pipeline|
|US20060259487 *||May 16, 2005||Nov 16, 2006||Microsoft Corporation||Creating secure process objects|
|US20070058807 *||Sep 15, 2005||Mar 15, 2007||Microsoft Corporation||Establishing a unique session key using a hardware functionality scan|
|US20080005438 *||Dec 30, 2004||Jan 3, 2008||Bin Xing||Methods and Apparatuses to Maintain Multiple Execution Contexts|
|US20090055623 *||Nov 4, 2008||Feb 26, 2009||International Business Machines Corporation||Method and Apparatus for Supporting Shared Library Text Replication Across a Fork System Call|
|US20090158036 *||Feb 23, 2009||Jun 18, 2009||Microsoft Corporation||protected computing environment|
|CN101309121B||May 18, 2007||Jan 23, 2013||希姆通信息技术（上海）有限公司||Method of realizing MUX protocol without need of adding drive of kernel under Linux platform|
|WO2006069484A1 *||Dec 30, 2004||Jul 6, 2006||Intel Corporation||Methods and apparatuses to maintain multiple execution contexts|
|U.S. Classification||718/104, 718/103|
|Jun 9, 1998||CC||Certificate of correction|
|Sep 4, 2001||REMI||Maintenance fee reminder mailed|
|Feb 11, 2002||LAPS||Lapse for failure to pay maintenance fees|
|Apr 9, 2002||FP||Expired due to failure to pay maintenance fee|
Effective date: 20020210